Tandem neural networks for designing nanoplasmonic structures to manipulate quantum dynamics

In the cutting-edge field of quantum nanophotonics, embedding quantum emitters (QEs) into nanostructures goes beyond the conventional exploration of light-structure interactions. The ability to optimize the local density of states (LDOS) through the design of nanophotonic structures for quantum manipulation is a significant area of research. Traditional inverse design methods are inefficient when dealing with complex structures and multiparameter optimizations, and there are issues such as the design non-uniqueness of the designed structures. In this work, we established a mapping between QEs and nanophotonic structures by introducing the LDOS and developed an artificial tandem neural network (TNN) to design nanophotonic structures with desired LDOS and optimize the spontaneous emission characteristics of QEs. The TNN enables precise one-to-one inverse design of nanostructures, significantly enhancing the LDOS at specific, desired frequency positions. Additionally, we introduce a structural loss function to address multiparameter optimizations, including the material thickness and types. Although our approach is exemplified by the design of two-layer core-shell structures that enhance the LDOS, it can also be readily extended to multilayer core-shell designs through transfer learning technology. Our work shows that TNN is an efficient methodology and a powerful tool for tailoring the LDOS and manipulating quantum dynamics in various nanophotonic structures.